opportunities for data analytics in power generation … · 2016. 7. 18. · opportunities for data...
TRANSCRIPT
![Page 1: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/1.jpg)
Opportunities for Data Analytics in Power Generation
Prognostics: The Final Frontier
Scott Affelt XMPLR Energy June 30, 2016
![Page 2: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/2.jpg)
XMPLR Energy • Consulting
– Business Strategy – Disruptive Technology Introduction – M&A
• Technology Liaison – Licensing & Technology Transfer – Foreign Market Introduction
• Data Analytics
![Page 3: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/3.jpg)
Key Opportunities for Data Analytics • Efficiency
– Fuel Costs – Capacity/Output
• Reliability – Availability – Capacity/Output – Load Factor – Maintenance
• Emissions – Compliance – Optimization
• Flexibility – Operational – Economic
Focus of Today’s Talk Data Science & Prognostics
![Page 4: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/4.jpg)
Why is this Important?
Source: Duke Energy. 2007-2012
![Page 5: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/5.jpg)
Why is this Important?
MIT Study: Bearing Failures in Rotating Equipment cause $240B in downtime and repair costs EVERY YEAR!
![Page 6: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/6.jpg)
Domain Expertise
Computer Science
Math & Statistics
How can Data Science Help?
Machine Learning
Data Processing
Traditional Software
DATA SCIENCE
![Page 7: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/7.jpg)
Approaches to Data Analytics
Difficulty
Value Descriptive Analytics
What happened?
Diagnostic Analytics
Why did it happen? Predictive
Analytics
When will it happen?
Prescriptive Analytics
What should I do about it?
Source: Gartner
![Page 8: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/8.jpg)
Diagnostics vs Prognostics
• Failure mode detection
• Fault Location • Detection • Isolation
• RUL Estimation • Degradation
prediction
• Health assessment • Severity detection • Degradation detection • Pattern Recognition
DIAGNOSTICS PROGNOSTICS
Source: IVHM Center
What Why
Where How
When
Phase 1 Diagnose
Phase 2 Predict
![Page 9: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/9.jpg)
Traditional Failure Curve
• Failure rate are high in the early life – “infant mortality” • Failure rates are high at the end of life • Remaining useful life can be predicted. • Based on assets that have regular & predictable wear • The Rise of Predictive Maintenance!
Source: EPRI
![Page 10: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/10.jpg)
Predictive Maintenance
Time
Equi
pmen
t He
alth
In
dex
Total Cost
Reactive: Run to Failure
Preventative: Schedule Base
Predictive: Condition-based Advanced Patterns
Prognostics
Cost
Fault Equipment Fails Here
Degradation Starts Here
Fault Occurs Here
![Page 11: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/11.jpg)
Predicting Failures is Hard Only 18% of Failures are age-related
82% of Failures are
random
Sources: RCM Guide, NASA, Sept. 2008, and U.S.. Navy Analysis of Submarine Maintenance Data 2006
„
![Page 12: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/12.jpg)
Using Predictive Analytics for Remaining Useful Life (RUL)
• Condition Monitoring – Identifies anomalies in data trends – Typically, provides little warning of a fault (hrs to days)
• Advanced Pattern Recognition – Specific signal patterns (features) linked to faults – Gives indication of likely future fault – Generally, more focused on diagnostics than prognostics
• Mean time to failure – Based on “average” life of particular asset – Needs large set of “crash test” data – But is your asset “average”?
• Prognostics – Predicting the time at which a system or a component will no longer perform its intended function – Includes a confidence level associated with the time prediction.
• (i.e. RUL of 3 months at 60% confidence level
![Page 13: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/13.jpg)
Predictive Analytics vs Prognostics Predictive Analytics
Prognostics
Something will happen at
some point
Explicit time window until failure with
confidence level
![Page 14: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/14.jpg)
Prognostic Approaches
Physics Based Models
Data-based Models Classification, fuzzy logic, NN,
regression
Experience Based Models Generic, statistical life, mean time to failure
Range of Applicability
Incr
easi
ng C
ost a
nd
Accu
racy
Experienced-based Models
Advantages • Based on actual failure experience • Rules-based • Simple • Little data required
Disadvantages • Little prediction capability • Requires subject matter experts • Needs continued observations • Difficult to scale to other assets
![Page 15: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/15.jpg)
Prognostic Approaches
Physics Based Models
Data-based Models Classification, fuzzy logic, NN,
regression
Experience Based Models Generic, statistical life, mean time to failure
Range of Applicability
Incr
easi
ng C
ost a
nd
Accu
racy
Data-based Models
Advantages • Relatively simple and fast to implement • Physical cause/effects understanding not
necessary • Gain understanding of physical behaviors
from large datasets • Good for complex processes
Disadvantages • Physical cause/effects not utilized • Difficult to balance generalizations and
specific learning trends • Requires large datasets to characterize fault
features
![Page 16: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/16.jpg)
Prognostic Approaches
Physics Based Models
Data-based Models Classification, fuzzy logic, NN,
regression
Experience Based Models Generic, statistical life, mean time to failure
Range of Applicability
Incr
easi
ng C
ost a
nd
Accu
racy
Physics-based Models
Advantages • Prediction results based on intuitive
cause/effects relationship • Calibration only needed for different cases • Drives sensor requirements • Accurate
Disadvantages • Model development may be hard • Requires complete knowledge of physical
process • Large datasets may be unpractical for real-
time predictions • Getting the “right” data may be difficult
![Page 17: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/17.jpg)
Prognostic Approaches
Physics Based Models
Data-based Models Classification, fuzzy logic, NN,
regression
Experience Based Models Generic, statistical life, mean time to failure
Range of Applicability
Incr
easi
ng C
ost a
nd
Accu
racy
Hybrid Models
Physics + Data-based
Hybrid Models • Use components of both Physics
and Data-based models • Can provide more robust and
accurate RULs • Can address disadvantages of each
type of model
![Page 18: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/18.jpg)
Power Generation Applications for Prognostics
Rotary Equipment • Turbines • Pumps • Generators • Compressors • Gearbox • Bearings • Fans
Other Possibilities • Boiler Tubes? • HRSG? • Condensers? • Heaters? • Valves?
![Page 19: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/19.jpg)
Prognostics Implementation Challenges
• Data – Measuring the right things – Cost of New Sensors – Access to multiple data sources – Clean data
• People – M&D centers: leverage expertise – In-house vs outsource
• Accuracy – Confidence in predictions – Uncertainty in predictions – Validation and verification
• Data Security – Various data sources – Cloud-based or local server – Sharing data/information within
company
![Page 20: Opportunities for Data Analytics in Power Generation … · 2016. 7. 18. · Opportunities for Data Analytics in Power Generation Prognostics: The Final Frontier Scott Affelt XMPLR](https://reader035.vdocuments.us/reader035/viewer/2022071114/5feb477139a63d03ce63815d/html5/thumbnails/20.jpg)
Conclusions • Data analytics can help operators manage and improve
reliability of generation assets.
• Prognostics can be used to determine a RUL with a time element and confidence level
• Operators can use the RUL to actively manage maintenance schedule and operating conditions in order to maintain reliability.